Investigation of The Performance of Yolov5 Models In The Detection and Classification of Shoulder Implants Using X-Ray Images

Elif Baykal Kablan

Makale detayları

Investigation of The Performance of Yolov5 Models In The Detection and Classification of Shoulder Implants Using X-Ray Images

Yazarlar:

Elif Baykal Kablan

Yayıncı:

Dokuz Eylul University

Ana Başlıklar:

Bilgisayar Görüşü ve Çoklu Ortam Hesaplama (Diğer)

Dergi:

Dokuz Eylül Üniversitesi Mühendislik Fakültesi Fen ve Mühendislik Dergisi Vol.27 Issue 79, pp.99-109

Belge Tipi:

Makale

Yayın Tarihi:

1/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001336025

pISSN:

1302-9304

Anahtar kelimeler:

Özet:

 Total Shoulder Arthroplasty TSA is a surgical procedure in which the damaged part of the shoulder is replaced with a prosthetic implant in patients with pain and restriction of movement due to damage to the shoulder joint. If prosthetic implants wear out or become damaged over time, prosthesis manufacturer and model information is needed, but this information can often be unclear due to deficiencies in medical records. Therefore, there is a need for automated systems to provide prosthesis information quickly and accurately, reducing reliance on specialists. In this study, an automatic system is proposed for the detection and classification of shoulder implant manufacturers using X-ray images. In the system, implant head region detection and classification processes were performed using YOLOv5n, YOLOv5s, YOLOv5m, and YOLOv5l models and the results obtained were compared. The class weighting technique was also used, where some classes are better represented than others due to data imbalance. It was found that this technique significantly improves the classification performance of the model. As a result of the comparison, it was observed that the YOLOv5l model with class weighting showed the highest performance in mAP0.5, mAP0.5:0.95, precision, sensitivity, and F-measure values, and these values were obtained as 98.3, 85.9, 97.0, 95.7 and 96.3 respectively. The proposed YOLOv5l model offers a low-cost and high-performance solution compared to other hybrid machine learning approaches in the literature. Thus, the effectiveness of YOLOv5 models as a valuable tool in the detection and classification of shoulder implants has been demonstrated. Furthermore, the proposed system is expected to provide efficient and reliable analysis processes and reduce the workload for radiologists and surgeons.

 Total Shoulder Arthroplasty TSA is a surgical procedure in which the damaged part of the shoulder is replaced with a prosthetic implant in patients with pain and restriction of movement due to damage to the shoulder joint. If prosthetic implants wear out or become damaged over time, prosthesis manufacturer and model information is needed, but this information can often be unclear due to deficiencies in medical records. Therefore, there is a need for automated systems to provide prosthesis information quickly and accurately, reducing reliance on specialists. In this study, an automatic system is proposed for the detection and classification of shoulder implant manufacturers using X-ray images. In the system, implant head region detection and classification processes were performed using YOLOv5n, YOLOv5s, YOLOv5m, and YOLOv5l models and the results obtained were compared. The class weighting technique was also used, where some classes are better represented than others due to data imbalance. It was found that this technique significantly improves the classification performance of the model. As a result of the comparison, it was observed that the YOLOv5l model with class weighting showed the highest performance in mAP0.5, mAP0.5:0.95, precision, sensitivity, and F-measure values, and these values were obtained as 98.3, 85.9, 97.0, 95.7 and 96.3 respectively. The proposed YOLOv5l model offers a low-cost and high-performance solution compared to other hybrid machine learning approaches in the literature. Thus, the effectiveness of YOLOv5 models as a valuable tool in the detection and classification of shoulder implants has been demonstrated. Furthermore, the proposed system is expected to provide efficient and reliable analysis processes and reduce the workload for radiologists and surgeons.

 Total Shoulder Arthroplasty TSA is a surgical procedure in which the damaged part of the shoulder is replaced with a prosthetic implant in patients with pain and restriction of movement due to damage to the shoulder joint. If prosthetic implants wear out or become damaged over time, prosthesis manufacturer and model information is needed, but this information can often be unclear due to deficiencies in medical records. Therefore, there is a need for automated systems to provide prosthesis information quickly and accurately, reducing reliance on specialists. In this study, an automatic system is proposed for the detection and classification of shoulder implant manufacturers using X-ray images. In the system, implant head region detection and classification processes were performed using YOLOv5n, YOLOv5s, YOLOv5m, and YOLOv5l models and the results obtained were compared. The class weighting technique was also used, where some classes are better represented than others due to data imbalance. It was found that this technique significantly improves the classification performance of the model. As a result of the comparison, it was observed that the YOLOv5l model with class weighting showed the highest performance in mAP0.5, mAP0.5:0.95, precision, sensitivity, and F-measure values, and these values were obtained as 98.3, 85.9, 97.0, 95.7 and 96.3 respectively. The proposed YOLOv5l model offers a low-cost and high-performance solution compared to other hybrid machine learning approaches in the literature. Thus, the effectiveness of YOLOv5 models as a valuable tool in the detection and classification of shoulder implants has been demonstrated. Furthermore, the proposed system is expected to provide efficient and reliable analysis processes and reduce the workload for radiologists and surgeons.

 Total Omuz Artroplastisi TOA, omuz eklemindeki hasara bağlı ağrı ve hareket kısıtlanması yaşayan hastalarda omuzun hasar görmüş kısmının protez implant ile değiştirildiği cerrahi işlemdir. Zamanla protez implantların aşınması veya zarar görmesi durumunda, protez üreticisi ve model bilgisine ihtiyaç duyulur, ancak tıbbi kayıtlardaki eksiklikler nedeniyle bu bilgiler genellikle belirsiz olabilmektedir. Bu nedenle, uzman kişiye bağımlılığı azaltarak hızlı ve doğru bir şekilde protez bilgilerini sağlayacak otomatik sistemlere ihtiyaç duyulmaktadır. Bu çalışmada, omuz implantı üreticilerinin X-ışını görüntüleri kullanılarak tespiti ve sınıflandırılması için otomatik bir sistem önerilmiştir. Sistemde YOLOv5n, YOLOv5s, YOLOv5m ve YOLOv5l modelleri kullanılarak implant baş bölgesi tespiti ve sınıflandırma işlemleri gerçekleştirilmiş ve elde edilen sonuçlar karşılaştırılmıştır. Veri dengesizliği nedeniyle bazı sınıfların diğerlerinden daha iyi temsil edildiği sınıf ağırlıklandırma tekniği de kullanılmıştır. Bu tekniğin modelin sınıflandırma performansını önemli ölçüde artırdığı tespit edilmiştir. Karşılaştırma sonucunda sınıf ağırlıklandırmalı YOLOv5l modelinin mAP0.5, mAP0.5:0.95, kesinlik, duyarlılık ve F1-ölçütü değerlerinde en yüksek performansı gösterdiği görülmüştür ve bu değerler sırasıyla 98.3, 85.9, 97.0, 95.7 ve 96.3 olarak elde edilmiştir. Önerilen YOLOv5l modeli, literatürdeki diğer hibrit makine öğrenmesi yaklaşımlarına göre düşük maliyetli ve yüksek performanslı bir çözüm sunmaktadır. Bu sayede, YOLOv5 modellerinin omuz implantlarının tespit ve sınıflandırmasında değerli bir araç olarak etkinliği gösterilmiştir. Ayrıca, önerilen sistem sayesinde radyologlar ve cerrahlar için verimli ve güvenilir analiz süreçleri sunulması ve iş yükünün azaltılması beklenmektedir.

Makale detayları

Investigation of The Performance of Yolov5 Models In The Detection and Classification of Shoulder Implants Using X-Ray Images

Yazarlar:

Elif Baykal Kablan

Yayıncı:

Dokuz Eylul University

Ana Başlıklar:

Bilgisayar Görüşü ve Çoklu Ortam Hesaplama (Diğer)

Dergi:

Dokuz Eylül Üniversitesi Mühendislik Fakültesi Fen ve Mühendislik Dergisi Vol.27 Issue 79, pp.99-109

Belge Tipi:

Makale

Yayın Tarihi:

1/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001336025

pISSN:

1302-9304

Deep Learning

Total Shoulder Arthroplasty

X-ray

YOLOv5Derin Öğrenme

Total Omuz Artroplastisi

X-ray

YOLOv5

Sayı

Makale detayları

Yazarlar: Elif Baykal Kablan

Yayıncı: Dokuz Eylul University

Ana Başlıklar: Bilgisayar Görüşü ve Çoklu Ortam Hesaplama (Diğer)

Dergi: Dokuz Eylül Üniversitesi Mühendislik Fakültesi Fen ve Mühendislik Dergisi Vol.27 Issue 79, pp.99-109

Belge Tipi: Makale

Yayın Tarihi: 1/2025

Belge Dili: Türkçe

Document ID (Org No.): 2001336025

pISSN: 1302-9304

Sayı

Yazarlar:

Elif Baykal Kablan

Yayıncı:

Dokuz Eylul University

Ana Başlıklar:

Bilgisayar Görüşü ve Çoklu Ortam Hesaplama (Diğer)

Dergi:

Dokuz Eylül Üniversitesi Mühendislik Fakültesi Fen ve Mühendislik Dergisi Vol.27 Issue 79, pp.99-109

Belge Tipi:

Makale

Yayın Tarihi:

1/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001336025

pISSN:

1302-9304